Application of Adaptive Kriging Method in Bridge Girder Reliability Analysis

Abstract

AbstractMonte Carlo Simulation (MCS) is typically employed to calculate the reliability of nonlinear limit states. Millions of trials are required in MCS to achieve a robust reliability calculation with an acceptable margin of error. The high computational cost of conducting millions of trails presents a major challenge for using MCS in reliability-based software’s. Modified methods have been proposed in the literature to reduce the computational cost of conducting reliability analyses for nonlinear limit states such as importance sampling, line sampling, subset simulation, etc. Although the existing modified methods reduce the computational cost, further optimization is required to improve the efficiency of the reliability calculation when used in commercial software packages. The objectives of this research are to: (1) develop a procedure for conducting Active Learning Kriging Monte Carlo Simulation (AK-MCS) analysis to assess the reliability of bridge girders; and (2) utilize the developed procedure in conducting a sensitivity analysis to propose an optimum Kriging configuration for the reliability analysis. The sensitivity analysis considered three key parameters in the Kriging part of the simulation: choice of regression model, correlation functions, and learning functions. A total of 486 analyses were conducted in this study. Analysis results recommend using quadratic regression function with a spline correlation function and H learning function to assess the reliability of bridge girders. Analysis results are limited to the considered bridge girders. Future research is recommended to propose a universal configuration for conducting AK-MCS for multiple bridge components including piers, girders behaving in composite action, bridge decks, abutments, etc.